from math import exp, log
from random import random
from typing import List

from gbso.program.test_case import Output, TestCase, eq_on_testcase
from gbso.program.mutate import mutate_program
from gbso.program.program import Program

EPSILON = 0.00001

DEFAULT_NUM_ITERS = 1_000_000

DEFAULT_PROB_OPCODE = 0.25
DEFAULT_PROB_OPERAND = 0.25
DEFAULT_PROB_SWAP = 0.25
DEFAULT_PROB_INSN = 0.25

DEFAULT_PROB_INSN_UNUSED = 0.1


def cost(orig_prgm, test_cases, outputs, prgm) -> int:
    c = prgm.perf() - orig_prgm.perf()
    # print(f"init cost: {c}")

    for test_case in test_cases:
        c += eq_on_testcase(orig_prgm, prgm, test_case, outputs)
        # print(f"cost after testcase: {c}")

    return c


def optimize(
    prgm: Program,
    max_size: int,
    test_cases: List[TestCase],
    outputs: List[Output],
    beta: int = 0.75,  # How far away in cost you are allowed to search
    num_iters: int = DEFAULT_NUM_ITERS,
    prob_opcode: float = DEFAULT_PROB_OPCODE,
    prob_operand: float = DEFAULT_PROB_OPERAND,
    prob_swap: float = DEFAULT_PROB_SWAP,
    prob_insn: float = DEFAULT_PROB_INSN,
    prob_insn_unused: float = DEFAULT_PROB_INSN_UNUSED,
) -> Program:
    padded_prgm = prgm.pad(max_size)

    last_prgm = padded_prgm
    last_cost = cost(padded_prgm, test_cases, outputs, last_prgm)

    for _ in range(num_iters):
        candidate_prgm = mutate_program(
            last_prgm, prob_opcode, prob_operand, prob_swap, prob_insn, prob_insn_unused
        )
        candidate_cost = cost(padded_prgm, test_cases, outputs, candidate_prgm)

        if candidate_cost < last_cost - log(random()) / beta:
            last_prgm = candidate_prgm
            last_cost = candidate_cost

    return last_prgm